Identification of predators from saliva traces on game species and/or livestock kills is gaining increasing importance in wildlife management, particularly in areas where direct wildlife-human conflicts regularly occur. When the noninvasive sampling of hairs and scats is difficult, as with rare and elusive predators, saliva samples constitute a potentially useful source of DNA. To test the feasibility of this approach in obtaining an accurate genotype of the predator, we applied an experimental approach. Captive wolves (Canis lupus) and lynxes (Lynx lynx) were allowed to feed on freshly killed roe deer (Capreolus capreolus) pieces for 1 min. After removal, pieces were sampled for saliva traces after 1, 24, and 48 h. Microsatellite analysis revealed that error rates and amplification failure increased sharply over time. While samples collected after 1 and 24 h yielded > 83% complete genotypes, values dropped to < 50% for samples collected after 48 h, of which 7% were incorrect even when consensus genotypes from 9 polymerase chain reactions were obtained. Our results stress the importance of rapid sampling after carcass detection, as well as implementing a multiple-tubes approach when using microsatellite markers for genetic predator identification based on saliva traces.